Article
Environmental Sciences
Masanori Onishi, Shuntaro Watanabe, Tadashi Nakashima, Takeshi Ise
Summary: This study evaluated the practicality and robustness of a tree identification system using UAVs and deep learning in temperate forests in Japan. The model showed high performance on dataset obtained at the same time and with the same tree crowns, but decreased performance on dataset from different times and sites. Misclassifications were observed between closely related species, species with similar leaf shapes, and species preferring the same environment. The study highlights the potential of using UAV RGB images and deep learning for practical tree identification.
Article
Environmental Sciences
Yifei Sun, Zhenbang Hao, Zhanbao Guo, Zhenhu Liu, Jiaxing Huang
Summary: This study explores the impact of sample distribution patterns on the accuracy and generalization performance of deep learning models for chestnut detection and classification. The results show that the combination of DeepLab V3 with ResNet-34 backbone performs the best, while the combination of DeepLab V3+ with ResNet-50 backbone performs the worst. Different spatial distribution patterns of chestnut planting also affect the classification accuracy. Comprehensive training data improves the generalization performance of chestnut classification with different spatial distribution patterns.
Review
Environmental Sciences
Yago Diez, Sarah Kentsch, Motohisa Fukuda, Maximo Larry Lopez Caceres, Koma Moritake, Mariano Cabezas
Summary: This article highlights the importance of forests to the planet and their current applications in forestry research, focusing on studies that use deep learning and RGB images collected by UAVs to address practical forestry research problems. The article summarizes the strengths and methodological issues of existing studies, and provides public data and code resources to aid researchers interested in working in this area.
Article
Computer Science, Information Systems
Passakron Phuangthongkham, Phonthep Angsuwatcharakon, Santi Kulpatcharapong, Peerapon Vateekul, Rungsun Rerknimitr
Summary: The determination of benign or malignant bile duct strictures is challenging, and current methods still lack consistency in diagnosis. Biopsy is commonly used for accurate diagnosis, but sampling errors can lead to false-negative results. In this study, a convolutional neural network is proposed for real-time classification of malignant biliary strictures. The model shows good performance in still images with sensitivity of 0.8577 and F1-score of 0.8395, and even better performance with video inference with sensitivity of 0.9024 and F1-score of 0.9193. The model can also achieve real-time inference at a speed of 83 frames per second.
Article
Agriculture, Multidisciplinary
Ong Win Kent, Tan Weng Chun, Tay Lee Choo, Lai Weng Kin
Summary: This study introduces an enhanced approach for early symptom detection of basal stem rot (BSR) disease in densely populated oil palm tree areas. The proposed method, utilizing a modified U-Net architecture and an image postprocessing method called the overlapped contour separation (OVCS) algorithm, demonstrates superior segmentation performance and accurate identification of tree boundaries.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2023)
Article
Agriculture, Multidisciplinary
Raphael Linker, Mery Dafny-Yalin
Summary: This study utilizes convolutional neural networks to identify visual symptoms of fire blight disease in pear trees and to detect autumn blooming. The trained CNNs are used to analyze images and generate infection/blooming maps, which can assist growers in managing orchards and reducing labor and economic costs.
PRECISION AGRICULTURE
(2023)
Article
Geosciences, Multidisciplinary
Biswajeet Pradhan, Saro Lee, Abhirup Dikshit, Hyesu Kim
Summary: Floods are natural disasters with severe impacts. Flood susceptibility maps can provide valuable information for reducing flood risks. This study proposes an explainable artificial intelligence (XAI) model, integrating the SHAP model, to interpret CNN deep learning models and analyze the variables influencing flood susceptibility mapping.
GEOSCIENCE FRONTIERS
(2023)
Article
Remote Sensing
Girma Tariku, Isabella Ghiglieno, Gianni Gilioli, Fulvio Gentilin, Stefano Armiraglio, Ivan Serina
Summary: This study presents a method that combines object-based supervised machine learning for dataset preparation and a pre-trained transfer learning model for precise plant species classification in heterogeneous areas. The test results show high classification accuracy, and a comparative study is conducted with other transfer learning models.
Article
Computer Science, Information Systems
Anupama Namburu, Prabha Selvaraj, Senthilkumar Mohan, Sumathi Ragavanantham, Elsayed Tag Eldin
Summary: Forest fires are caused by natural factors like lightning, high temperatures, and dryness. India has experienced an increase in the frequency of forest fires, with 136,604 fire points detected between January and March 2022. While satellite monitoring provides valuable information, video-based fire detection on the ground using unmanned aerial vehicles equipped with high-resolution cameras can identify fires more quickly. This paper proposes a cheaper UAV with deep learning capabilities to classify forest fires (97.26%) and share the detection and GPS location with state forest departments.
Article
Agriculture, Multidisciplinary
Chong He, Yongliang Qiao, Rui Mao, Mei Li, Meili Wang
Summary: This study proposes a sheep live weight estimation approach based on LiteHRNet using RGB-D images. Experimental results show that the lightweight CNN model trained on RGB-D images can achieve acceptable weight estimation results, with a Mean Average Percentage Error (MAPE) of 14.605% and only 1.06M parameters. The results of this study have the potential to develop an embedded device for automatic sheep live weight estimation and contribute to the development of precision livestock farming.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2023)
Article
Remote Sensing
Joao Valente, Santosh Hiremath, Mar Ariza-Sentis, Marty Doldersum, Lammert Kooistra
Summary: This study investigated the feasibility of using aerial images from unmanned aerial vehicles (UAV) and deep learning to map Rumex in grasslands. Results showed that the detection of Rumex was highly dependent on the flight height, with the MobileNet model performing best at 10 meters.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Article
Environmental Sciences
Joseph K. Mhango, Edwin W. Harris, Richard Green, James M. Monaghan
Summary: This study utilized the Faster Region-based Convolutional Neural Network (FRCNN) framework to produce a plant detection model and estimate plant densities using UAV imagery, showing the accurate construction of two-dimensional maps of plant density with high correlation to important yield components. Despite the challenges of inaccurate predictions in images of merged canopies, the FRCNN model proved to be effective in predicting plant density and its relationship with potato yield attributes.
Article
Environmental Sciences
Binbin Wang, Guijun Yang, Hao Yang, Jinan Gu, Sizhe Xu, Dan Zhao, Bo Xu
Summary: This study utilizes UAV remote sensing technology and a deep learning algorithm to improve the detection and counting of maize tassels, addressing the challenges posed by the complex field environment. The improved RetinaNet model demonstrates significant advancements in accuracy compared to other mainstream target detection models.
Article
Computer Science, Interdisciplinary Applications
Seyma Akca, Nizar Polat
Summary: Field inspection of tree counts in orchards is time-consuming, but using imaging systems integrated with UAVs and deep learning algorithms can provide a more efficient and accurate method. The study presented a CNN architecture for semantic segmentation of trees, shadows, and soil in orchards using high-resolution orthophoto. The results showed high recall and precision rates, indicating the effectiveness of the proposed method.
EARTH SCIENCE INFORMATICS
(2022)
Article
Plant Sciences
Ling Zheng, Mingyue Zhao, Jinchen Zhu, Linsheng Huang, Jinling Zhao, Dong Liang, Dongyan Zhang
Summary: In this study, a method for identification of soybean kernel damages was developed by combining HSI and RGB images and improved ShuffleNet. The proposed method achieved high recognition accuracy and reliability, and can be extended to analyze other quality indicators of crop kernels.
FRONTIERS IN PLANT SCIENCE
(2023)